Scaling after a pilot: cohorts, cancellations and cost

The post-pilot decision: increase, maintain or stop

A successful pilot does not prove that doubling the budget will preserve cost per outcome. The next traffic increment may differ by audience, placement, device and processing time. The advertiser's task is to grow verified volume while seeing the price of an additional result and retaining time to correct deterioration. This guide produces a decision record for one source, not a promise of a universal growth rate.

The illustrative scenario is in Brazil: an imaginary service helps small stores manage a catalog. It is not a live AffiliateCore offer. Before a real test, check product eligibility, service geography, support language and permitted sources. Every figure below is invented solely to teach calculation, not to describe a client or market.

You need completed pilot results, an agreed target-action definition, an expense ledger, lead statuses and an owner for processing. If leads remain unreviewed or sources cannot be distinguished, resolve that gap first. More budget does not fix unknown economics.

1. Define the unit of comparison

Compare cohorts: groups of leads with the same inclusion rule and observation age. Start with submission week, country, source and offer version. Do not split a small sample into dozens of cells; one confirmed lead cannot reliably rank placements.

Record submission time separately from the latest status-change time. A lead confirmed today may have arrived last week; it does not belong automatically with today's inquiries. For each cohort, retain pending, confirmed, rejected and canceled-after-confirmation states. Check whether your system's statuses are mutually exclusive. Historical confirmations and their later cancellations are not separate leads to add together.

Use the prepared record when registering as an AffiliateCore advertiser: describe your own offer and outcome criteria. Complete applicable account requirements and prepare an offer draft. Listing alone guarantees neither approval, partners nor leads. Existing advertisers can sign in.

2. Wait for comparable data maturity

Choose a review window based on your product's actual process: 14 days after submission is an illustrative assumption only. Comparing a 14-day-old cohort with yesterday's leads is invalid. Keep an operational report for recent inquiries and a separate report for sufficiently reviewed outcomes.

The observation window is an analytical rule, not a mandatory platform funds hold. Separate lead verification, outcome cancellation and payout timing. Include repeat purchases only when events can actually be linked to the original customer and the outcome definition allows them. Do not invent future revenue to justify a losing test.

Illustrative decision route: new cohort → equal observation age → reconcile statuses and costs → check processing capacity → bounded increase → another comparable cohort. Insufficient data sends the decision back to observation, not automatically to expansion.

3. Calculate cost per retained outcome

Illustrative source A produces 100 leads. After the agreed window, 60 have been confirmed historically, including 10 subsequently canceled: retained confirmations are 60 − 10 = 50. Costs are 1,000 currency units for placement, 200 for materials and 100 for processing. Total: 1,300; cost per retained outcome: 1,300/50 = 26.

Source B produces 80 leads, 48 historical confirmations and 4 cancellations, leaving 44 retained outcomes. Total costs are 900 + 150 + 50 = 1,100; cost per retained outcome is 1,100/44 = 25. Cost per incoming lead is 13 for A and 13.75 for B. A looks cheaper at submission; B looks slightly cheaper after verification. This is arithmetic, not evidence of statistically significant superiority.

Count actual partner payments, fees, production and servicing without double counting. If placement is already included in the partner rate, do not add it again. Explain how you allocate one-time expenses and separate variable costs. Use the same method, currency and period for both sources.

4. Inspect the marginal cost of expansion

The old average conceals the price of added volume. Suppose the baseline cohort delivers 50 retained outcomes for 1,300. After a change, a comparable new cohort delivers 65 for 1,950. The differences are 650 in cost and 15 outcomes: 650/15 ≈ 43.33 per additional outcome. The new average is 1,950/65 = 30.

This does not prove causal lift from the budget: seasonality and other changes could affect both numbers. It identifies a question to investigate. If the illustrative economic ceiling is 35, an average of 30 still fits, while the observed additional-outcome cost of 43.33 needs investigation. Do not replace a marginal assessment with an attractive average.

Where feasible, retain a comparable control group and predefine allocation to investigate causality. Otherwise label the comparison observational and list differences between periods. Attribution reporting alone does not establish incrementality.

5. Increase one variable

Choose one source, one country and one offer version. Predefine a bounded volume increase, spending ceiling, evaluation period and owner. Step size depends on your financial and operational ability to absorb a failed test; there is no universally safe percentage.

Do not simultaneously change the rate, landing page, acceptance criteria and placement mix. If changes are unavoidable, introduce a new terms version and acknowledge that the comparison is no longer clean. Do not blame source quality merely because your team could not process the larger queue.

Before increasing, check form availability, response time, queue capacity and ownership of disputed cases. If a team processes 30 inquiries per day, increasing to 60 without added capacity changes the process itself. Delayed handling can reduce confirmation independently of visitor quality.

6. Define stopping rules and communication

Pause expansion immediately for broken event tracking, unavailable landing pages, country mismatches or data exposure. Tie an economic stop to the predefined spending limit and sufficiently mature outcomes. A temporary increase in pending leads does not prove that a source has deteriorated.

Illustrative partner notice: “From the stated date we will test a new limit for this country and offer version. The rate and acceptance criterion are specified separately. Earlier events will be reviewed under the applicable previous terms. The next reconciliation follows the agreed observation window; direct questions to the owner.” This is a drafting template, not permission to send messages.

Do not change conditions retrospectively. Specify the effective time and timezone, handling of previously acquired events and a dispute contact. First agree actual terms through the platform's applicable process, then update partner materials.

7. Decision record and exercise

The record includes source, country, terms version, cohort rule, window, expense categories, incoming leads, current statuses, retained confirmations, unit cost, operational load and unknowns. Conclude “increase within a limit,” “maintain volume,” “fix the process” or “stop,” with a reason and next evaluation date.

Before deciding, check equal cohort age, no double counting, included cancellations, complete costs, a control or explicit comparison limitations, sufficient handling capacity, clear partner terms and no profitability promises. Retain original exports with the minimum permitted data and the calculation version so a dispute can be checked.

Exercise: complete records for two sources, reproduce the calculation above, then define one change and one stopping limit. Prepare your own AffiliateCore offer when starting; existing advertisers use their account and applicable review process. Do not create fictional offers in the live catalog for this exercise.